Convergent Use of Phosphatidic Acid for Hepatitis C Virus and SARS-Cov

bioRxiv preprint doi: https://doi.org/10.1101/2021.05.10.443480; this version posted May 10, 2021. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

1 2 Convergent use of phosphatidic acid for Hepatitis C virus and 3 SARS-CoV-2 replication organelle formation

4 5 Keisuke Tabata1†* , Vibhu Prasad1*, David Paul1‡, Ji-Young Lee1, Minh-Tu Pham1, Woan-Ing 6 Twu1, Christopher J. Neufeldt1, Mirko Cortese1, Berati Cerikan1, Cong Si Tran1, Christian 7 Lüchtenborg2, Philip V’kovski3,4, Katrin Hörmann5, André C. Müller5, Carolin Zitzmann6‼, 8 Uta Haselmann1, Jürgen Beneke7, Lars Kaderali6, Holger Erfle7, Volker Thiel3,4, Volker 9 Lohmann1, Giulio Superti-Furga5,8, Britta Brügger2, and Ralf Bartenschlager1,9,10, & 10 11 Affiliations: 12 1Department of Infectious Diseases, Molecular Virology, Heidelberg University, 13 Heidelberg, Germany 14 2Biochemistry Center Heidelberg, Heidelberg University, Heidelberg, Germany 15 3Institute of Virology and Immunology IVI, Bern, Switzerland. 16 4Department of Infectious Diseases and Pathobiology, Vetsuisse Faculty, University of 17 Bern, Bern, Switzerland. 18 5CeMM Research Center for Molecular Medicine of the Austrian Academy of Sciences, 19 Vienna, Austria. 20 6Institute of Bioinformatics and Center for Functional Genomics of Microbes, University 21 Medicine Greifswald, Greifswald, Germany. 22 7BioQuant, Heidelberg University, Heidelberg, Germany. 23 8Center for Physiology and Pharmacology, Medical University of Vienna, Vienna, Austria. 24 9Division Virus-Associated Carcinogenesis, German Cancer Research Center, Heidelberg, 25 Germany 26 10German Center for Infection Research, Heidelberg Partner Site, Heidelberg, Germany 27 28 †Keisuke Tabata: Department of Genetics, Graduate School of Medicine, Osaka University, 29 Osaka, Japan; Laboratory of Intracellular Membrane Dynamics, Graduate School of Frontier 30 Biosciences, Osaka University 31 ‡David Paul: MRC Laboratory of Molecular Biology, Francis Crick Avenue, Cambridge 32 CB2 0QH, UK 33 ‼Carolin Zitzmann: Los Alamos National Laboratory, Theoretical Biology and Biophysics 34 Los Alamos, NM, USA 35 36 * These authors contributed equally to this work 37 38 & Correspondence: 39 Ralf Bartenschlager: [email protected] ; Phone: +49-6221- 40 564225; Fax: +49-6221-564570 41

42 Abstract 43 Double membrane vesicles (DMVs) are used as replication organelles by phylogenetically 44 and biologically distant pathogenic RNA viruses such as hepatitis C virus (HCV) and severe 45 acute respiratory syndrome coronavirus-2 (SARS-CoV-2). Viral DMVs are morphologically 46 analogous to DMVs formed during autophagy, and although the proteins required for DMV 47 formation are extensively studied, the lipids driving their biogenesis are largely unknown. 48 Here we show that production of the lipid phosphatidic acid (PA) by acylglycerolphosphate 49 acyltransferase (AGPAT) 1 and 2 in the ER is important for DMV biogenesis in viral 50 replication and autophagy. Using DMVs in HCV-replicating cells as model, we found that 51 AGPATs are recruited to and critically contribute to HCV replication and DMV formation. 52 AGPAT1/2 double knockout also impaired SARS-CoV-2 replication and the formation of 53 autophagosome-like structures. By using correlative light and electron microscopy, we 54 observed the relocalization of AGPAT proteins to HCV and SARS-CoV-2 induced DMVs. In 55 addition, an intracellular PA sensor accumulated at viral DMV formation sites, consistent 56 with elevated levels of PA in fractions of purified DMVs analyzed by lipidomics. Apart from 57 AGPATs, PA is generated by alternative pathways via phosphotidylcholine (PC) and 58 diacylglycerol (DAG). Pharmacological inhibition of these synthesis pathways also impaired 59 HCV and SARS-CoV-2 replication as well as formation of autophagosome-like DMVs. 60 These data identify PA as an important lipid used for replication organelle formation by HCV 61 and SARS-CoV-2, two phylogenetically disparate viruses causing very different diseases, i.e. 62 chronic liver disease and COVID-19, respectively. In addition, our data argue that host- 63 targeting therapy aiming at PA synthesis pathways might be suitable to attenuate replication 64 of these viruses. 65

66 One Sentence Summary 67 Phosphatidic acid is important for the formation of double membrane vesicles, serving as 68 replication organelles of hepatitis C virus and SARS-CoV-2, and offering a possible host- 69 targeting strategy to treat SARS-CoV-2 infection. 70 71 72 Main Text 73 Chronic hepatitis C and COVID-19 are major medical problems. Both diseases are 74 caused by viral infections inflicting a large number of people and having led to millions of 75 deaths 1, 2. Chronic hepatitis C is caused by persistent infection with the hepatitis C virus 76 (HCV), while COVID-19 is due to acute infection with the severe acute respiratory syndrome 77 coronavirus-2 (SARS-CoV-2). Both viruses are biologically very distinct e.g. by having a 78 very narrow tropism and a predominantly persistent course of infection in the case of HCV, 79 contrasting the rather broad tropism and acute self-limiting course of infection in the case of 80 SARS-CoV-2. This biological distinction is reflected by their phylogenetic distance with 81 HCV belonging to the Flaviviridae and SARS-CoV-2 being a member of the Coronaviridae 82 virus family 3. In spite of these differences, both viruses possess a single strand RNA genome 83 of positive polarity that is replicated in membranous vesicles in the cytoplasm of infected 84 cells 4, 5. These vesicles are induced by viral proteins, in concert with cellular factors, and 85 composed of two membrane bilayers, thus corresponding to double-membrane vesicles 86 (DMVs). These DMVs accumulate in infected cells and can be regarded as viral replication 87 organelle. Viral DMVs have morphological similarity to autophagosomes 6, 7, but while 88 autophagy-induced DMVs serve to engulf cellular content and damaged organelles for 89 subsequent degradation, viral DMVs create a conducive and protective environment for 90 productive viral RNA replication. In the case of HCV and SARS-CoV-2, DMVs are derived

91 from the ER 8, 9, 10 and can be induced by the nonstructural proteins (NS)3, 4A, 4B, 5A and 92 5B in the case of HCV 7 and the viral proteins nsp3-4 in the case of MERS-CoV and SARS- 93 CoV 11, 12, alongside with co-opted host cell proteins and lipids. Here, we set-out to search for 94 common host cell factors exploited by the phylogenetically distant HCV and SARS-CoV-2 to 95 build up their cytoplasmic replication organelle. 96 Using HCV as a model to study DMV biogenesis, we purified DMVs under native 97 conditions and determined their molecular composition by proteomic profiling (Fig. 1A and 98 B). To this end we used human hepatoma cells (Huh7) containing a self-replicating HCV 99 replicon RNA (designated sg4BHA31R; 13) in which NS4B was HA-tagged (fig. S1A). This 100 RNA replicates autonomously and induces an extensive array of DMVs that can be isolated 101 by HA-affinity purification 13. Mass spectrometry-based proteomics analysis identified a total 102 of 1487 proteins significantly enriched in the NS4B-HA sample relative to the untagged 103 technical negative control (using SAINT average P-values >0.95) (data S1). Label free 104 quantitation (LFQ) revealed a major overlap of proteins (1542) between the NS4B-HA 105 complex and HCV-naïve ER membranes purified in parallel from Huh7 cells stably 106 expressing HA-tagged Calnexin (CNX-HA) (Fig. 1B and fig. S1B). Of note, 309 proteins 107 were significantly enriched in the NS4B-HA sample relative to the ER control with an over- 108 representation of proteins involved in RNA metabolism, intracellular vesicle organization and 109 transport as well as endomembrane organization (fig. S2). Given our interest in identifying 110 proteins of relevance for DMV formation, we selected 139 candidates with a bias for proteins 111 involved in vesicle transport and biogenesis as well as lipid metabolism. These candidates 112 were validated with respect to their role in HCV replication by using RNA interference-based 113 screening (Fig. 1C and data S2). In this way we could validate 38 hits as HCV dependency 114 factors. Amongst identified hits were acylglycerolphosphate acyltransferase (AGPAT) 1 and 115 2, two enzymes that catalyze the de novo formation of phosphatidic acid (PA), a precursor to 116 di- and triacylglycerols as well as all glycerophospholipids 14, 15. In addition, PA is involved 117 in signaling and protein recruitment to membranes and, owing to its small and highly charged 118 head group, promotes membrane curvature 16, 17, 18. Since these properties might be involved 119 in DMV formation, we focused our subsequent analysis on AGPATs. 120 AGPATs play crucial roles in lipid homeostasis, because enzyme-inactivating mutations 121 in AGPAT2 are linked to congenital generalized lipodystrophy and defects in PA metabolism 122 as well as autophagy are associated with neurological disorders and chronic obstructive 123 pulmonary disease 18, 19. Moreover, severe lipodystrophy as well as extreme insulin resistance 124 and hepatic steatosis have been observed in AGPAT2-/- mice 14. To date, 11 AGPATs have 125 been identified in mammalian cells. AGPAT1 to 5 preferentially utilize lysophosphatidic acid 126 (LPA) as an acyl donor while AGPAT6 to 11 preferentially utilize alternative 127 lysophospholipid substrates or have a preference for glycerol-3-phosphate. Thus, only 128 AGPAT1 to 5 function as true LPA acyltransferases 14. To establish which AGPAT family 129 members are found in NS4B-associated membranes, FLAG-tagged versions of each of the 5 130 AGPATs were transiently expressed in cells containing the HCV replicon sg4BHA31R (fig. 131 S3A). Pull-down of NS4B-HA revealed association with AGPAT1 and 2, and to a lesser 132 extent with AGPAT3, but not with AGPAT4 and 5. Additionally, endogenous AGPAT1 and 133 2 were detected in NS4B-HA containing membranes isolated from replicon-containing cells 134 (Fig. 1D), whereas AGPAT 3 was not enriched. Moreover, in HCV infected cells AGPAT1 135 and 2 were recruited to NS4B-containing sites that most likely correspond to sites of DMV 136 accumulation 13 (Fig. 1E). 137 To validate the role of AGPAT1 and 2 in HCV replication, we created knock-out cells 138 using CRISPR/Cas9. Although we observed reduced cell growth of stable double knock-out 139 (DKO) cells 8 days after transduction of guide RNAs, single KO cell pools showed no 140 decrease in cell growth and could be used for transient knock-out of the other AGPAT gene 141 without impacting cell viability for up to 8 days after transduction (fig. S3B). Using this 142 approach, we observed that AGPAT1/2 DKO impaired lipid droplet formation (fig. S3, C to

143 E) as shown previously 20, 21, confirming disruption of AGPAT1/2 function. To monitor the 144 impact of single KO and AGPAT1/2 DKO on HCV replication, cells were infected with an 145 HCV reporter virus and viral replication was determined by using luciferase assay. While 146 single KO suppressed HCV replication by ~50-70%, a reduction by ~90% was observed in 147 DKO cells (Fig. 1F). Even stronger replication suppression was observed with a subgenomic 148 replicon (fig. S4A), confirming that AGPAT depletion affected viral RNA replication and not 149 virus entry or assembly. Of note, replication was completely restored by stable expression of 150 AGPAT1 and 2 in DKO cells, which was not the case with either or both enzymatically 151 inactive mutants (Fig. 1G). In contrast, replication of Dengue virus (DENV) and Zika virus 152 (ZIKV), also belonging to the Flaviviridae family, but inducing morphologically different 153 membrane alterations, i.e. ER membrane invaginations 4, was not affected as determined by 154 plaque assay or with a reporter virus (Fig. 1H and fig. S4B, respectively). These results 155 suggest that enzymatically active AGPAT1 and 2 are required for HCV replication with both 156 AGPATs having partially redundant functions. 157 Next, we determined the impact of AGPAT KO on HCV-induced DMV formation. Since 158 AGPAT1/2 DKO reduces RNA replication, we employed a replication-independent system in 159 which DMV production is induced by the sole expression of an HCV NS3-5B polyprotein 160 fragment that undergoes self-cleavage to produce functional NS3, 4A, 4B, 5A and 5B 8, 22 161 (Fig. 2A). To determine the replicase subcellular location by fluorescence microscopy, NS5A 162 was fluorescently tagged with EGFP. This tagging has no effect on replicase functionality 8, 22. 163 While expression of this polyprotein induced a high number of DMVs in control cells, DMV 164 abundance was dramatically reduced in AGPAT1/2 DKO cells (Fig. 2, A and B), although 165 amounts of viral proteins were comparable in control and DKO cell pools (Fig. 2C). 166 Moreover, DMVs had a smaller diameter in AGPAT2 KO cells (fig. S4C). These results 167 argue for a pivotal role of AGPATs in HCV DMV biogenesis. 168 Given that AGPAT1 and 2 are important for DMV formation and their enzymatic 169 activity is required for HCV replication, we next focused on their reaction product, i.e. the 170 lipid PA. To quantify the amount of PA associated with HCV-induced DMVs and compare it 171 to ER membranes, we determined the lipidome of highly purified DMVs isolated from cells 172 containing the sg4BHA31R replicon (Fig. 2D). Consistent with earlier results, these 173 membranes contained elevated amounts of cholesterol and sphingolipids, which served as 174 positive controls, relative to ER membranes purified in parallel 13, 23. Of note, PA abundance 175 in DMVs also was increased in comparison to ER membranes, whereas the level of diacyl 176 phosphatidylcholine (aPC) and several other lipids was not affected (Fig. 2D; for further 177 lipids see data S3). 178 To confirm these findings in single cells, we used two alternative methods to detect PA 179 by fluorescence microscopy. First, we generated a recombinant protein composed of GST 180 that was fused to the PA binding domain (PABD) derived from yeast Spo20p (fig. S5A and 181 B). As a specificity control we employed the analogous sensor protein containing a mutation 182 in the PABD that abolishes PA binding, and GST alone 24. These proteins were introduced 183 via transient permeabilization into Huh7 derived cells (fig. S5C). In cells treated with phorbol 184 12-myristate 13-acetate (PMA), a potent activator of phospholipase D-mediated PA 185 production, as expected the intact sensor predominantly stained the plasma membrane, which 186 was not the case with the PA non-binding mutant or GST alone, confirming specificity of the 187 signal (fig. S5D). Moreover, also in cells that were not treated with PMA, the PA sensor 188 predominantly stained the plasma membrane (fig. S5D, right panel). Using this assay, we 189 monitored intracellular PA distribution in HCV replicon-containing cells and observed PA 190 colocalization with NS4B (fig. S5E). As second assay for intracellular PA detection, we 191 created a GFP-tagged sensor fused to the PABD of Raf1, a serine-threonine kinase recruited 192 to cellular membranes via its interaction with Ras and PA 25. While in control Huh7 cells this 193 PA sensor displayed a diffuse pattern (fig. S6A), upon co-expression of the HCV NS3-5B 194 polyprotein the sensor accumulated in NS5A-positive puncta (Fig. 2E). Of note, a control PA

195 sensor containing mutations in the PABD of Raf1 (mutant 4E) 26 displayed only a diffuse 196 pattern in NS3-5B expressing cells (Fig. 2E), supporting specificity of the signal and PA 197 recruitment to HCV replication sites. 198 Since these data suggest an important role of AGPAT1 and 2-dependent PA 199 enrichment on HCV-induced DMVs, we hypothesized that other pathways contributing to PA 200 generation in cells might also play a role in HCV replication. Apart from AGPATs, one other 201 route for PA synthesis is through hydrolysis of phosphatidylcholine (PC) by phospholipase 202 D1 (PLD1) and D2 (PLD2) enzymes (Fig. 2F, top panel) 17, 27. To test the role of PLD1/2 203 enzymes in HCV replication, we employed a pharmacological approach using 3 different 204 PLD1/2 inhibitors. Treatment with PLD2 inhibitor ML298 caused replication inhibition at a 205 concentration that did not significantly reduce cell viability (~25 µM; Fig. 2F, bottom panel), 206 whereas for the other drugs the reduction in HCV replication correlated with cytotoxicity (not 207 shown). In summary, these results suggest that PA generated via AGPAT1/2, and possibly by 208 alternative PA synthesis pathway, contributes to HCV replication by supporting the formation 209 of DMVs, which is the site of viral RNA amplification. 210 Virus-induced DMVs are morphologically analogous to autophagosomes generated 211 during autophagy 7; therefore, we tested if PA would be recruited to and is required for 212 autophagy-induced DMVs. To this end, we monitored the localization of the GFP tagged PA 213 sensor with markers for DMVs induced during nonselective and selective autophagy. To 214 monitor DMV formation induced during nonselective autophagy, cells were incubated in 215 starvation medium with or without bafilomycin A1 (BafA1), an inhibitor of the vacuolar-type 216 H+-ATPase inducing the accumulation of LC3-positive puncta, which are indicative of 217 autophagosomes. For selective autophagy events, we focused on the induction of DMVs 218 during mitophagy induced by treatment of the cells with valinomycin (Val) 28, 29. As shown in 219 Fig. 2G (top row), the PA sensor GFP-PABD-Raf1 was rather uniformly distributed 220 throughout the cell in non-induced cells. However, induction of nonselective autophagy by 221 serum starvation led to a significant increase in the number of LC3 puncta with GFP-PABD- 222 Raf1 relocalizing to these puncta (Fig. 2G). Similarly, induction of mitophagy by Val 223 treatment caused an abundant association of mCherry-Parkin puncta with GFP-PABD-Raf1 224 (Fig. 2G, lower panel), whereas in control cells not treated with Val, no such association was 225 found (fig. S6B). Next, we investigated the functional role of PA generation during 226 nonselective and selective autophagy. Consistent with the relocalization of PA to LC3 puncta 227 during nonselective autophagy, PA inhibitors targeting PLD1, PLD2 and AGPATs, applied 228 as short-term treatments and at non-toxic concentrations, significantly reduced the 229 accumulation of LC3 puncta (fig. S7). These findings are consistent with a recent study 230 suggesting that PA generated on the ATG16L1-positive autophagosome precursor membrane 231 contributes to autophagosome formation 30. Of note, a third pathway for PA production via 232 phosphorylation of diacylglycerol (DAG) by diacylglycerol kinase (DAGK) 27, did not 233 contribute to PA accumulation or increase in LC3 puncta during nonselective autophagy (fig. 234 S7). 235 Having found that AGPAT1 and 2, and their reaction product PA, are involved in DMV 236 formation induced upon HCV infection and in, morphologically similar, DMVs generated 237 during autophagy, we hypothesized that AGPATs and PA might also be involved in the 238 biogenesis of replication organelles of other unrelated RNA viruses, e.g., coronaviruses, 239 which also utilize DMVs as viral replication sites 9, 10. Hence, we investigated the role of 240 AGPATs in the DMV biogenesis of SARS-CoV-2, the causative agent of the ongoing 241 COVID-19 pandemic. In the first set of experiments, we studied the recruitment of AGPATs 242 to SARS-CoV-2 induced DMVs. In the case of MERS-CoV and SARS-CoV, formation of 243 DMVs with structural resemblance to those observed in infected cells can be induced by the 244 sole expression of viral nonstructural protein (nsp)3-4, which is an ~270 kilodalton large 245 polyprotein fragment undergoing self-cleavage 12. Building on these results we first 246 determined whether the same applies to SARS-CoV-2. Huh7-derived cells stably expressing

247 T7 RNA polymerase were transiently transfected with a T7 promoter driven SARS-CoV-2 248 HA-nsp3-4-V5 expression construct or the empty vector (fig S8A). Using 249 immunofluoresence with an HA-specific antibody in many cells we observed clusters of HA- 250 nsp3 (fig. S8B). Western blotting confirmed efficient self-cleavage between nsp3 and nsp4 251 (fig. S8C). To identify membrane alterations in HA-nsp3-4-V5 expressing cells, we 252 employed CLEM. Cells were transfected with the analogous expression construct encoding in 253 addition the NeonGreen gene to allow visualization of transfected cells by fluorescence 254 microscopy (fig. S8D). NeonGreen positive cells were recorded and examined by 255 transmission electron microscopy, revealing abundant clusters of DMVs (fig S8D). 256 Comparison of DMVs induced by nsp3-4 expression and by SARS-CoV-2 infection revealed 257 similar morphology, although expression-induced DMVs were smaller (~125 nm compared 258 to ~300 nm, respectively) (fig S8E). These results show that the sole expression of SARS- 259 CoV-2 nsp3-4 is sufficient to induce DMVs with structural similarity to those generated in 260 infected cells. 261 Next, we employed this expression-based system to determine AGPAT function in 262 SARS-CoV-2 nsp3-4 induced DMV formation. Huh7-derived cells expressing GFP-tagged 263 AGPAT1 or 2 were transiently transfected with the SARS-CoV-2 HA-nsp3-4-V5 encoding 264 plasmid or the empty vector and colocalization of AGPATs with HA-nsp3 was determined by 265 immunofluorescence microscopy. While in empty vector-transfected cells AGPAT2 and 1 266 were homogeneously distributed throughout the ER (Fig. 3A and fig. S9A, respectively), we 267 observed a strong relocalization of AGPATs in HA-nsp3-4-V5 expressing cells with 268 AGPATs forming puncta that colocalized with HA-nsp3 (Fig. 3, A and B; fig. S9A). Of note, 269 the relocalization of AGPATs induced by HA-nsp3-4-V5 was not the result of the massive 270 ER alterations occurring in SARS-CoV-2 infected cells, since the subcellular distribution of 271 other ER resident proteins, such as protein disulfide-isomerase (PDI) and calnexin remained 272 unaffected compared to the large puncta observed with AGPATs (Fig. 3C). Since SARS- 273 CoV-2 replication organelles are comprised of DMVs, convoluted membranes and zippered 274 ER 31, we next investigated the membrane structures at the sites of AGPAT colocalization 275 with HA-nsp3-4-V5. Using correlative light electron microscopy, we found that relocalized 276 AGPAT puncta perfectly correlated with extensive networks of SARS-CoV-2 HA-nsp3-4-V5 277 induced DMVs (Fig. 3D). Overall, the data shown here suggest that similar to HCV, 278 AGPATs are relocalized to SARS-CoV-2 nsp3-4 induced DMVs, the likely sites of viral 279 RNA replication 32. 280 Next, we tested the effect of AGPAT1/2 depletion on SARS-CoV-2 infection and 281 replication. To this end we used DKO Huh7-Lunet/T7 cells that were employed for the 282 imaging analyses described so far and stably introduced the SARS-CoV-2 receptor gene 283 ACE2. Viral replication was measured by using an image-based assay that quantifies the 284 number of cells containing detectable amounts of the nucleocapsid (N) protein (fig. S9B). 285 Using this approach, we observed significant reduction of SARS-CoV-2 positive cells in both 286 single and double AGPAT knockout cells (Fig. 3E). Consistently, RT-qPCR revealed similar 287 reduction of viral replication in single and double KO cells (Fig. 3E, lower right panel). To 288 determine if reduced SARS-CoV-2 replication in AGPAT1/2 KO cells might correlate with 289 altered DMV formation, we transiently expressed SARS-CoV-2 HA-nsp3-4-V5 in control, 290 single and double KO cells. The absence of AGPAT 1/2 did not significantly affect the 291 abundance of cleaved viral proteins HA-nsp3 and nsp4-V5 (fig. S8C). EM analysis of control 292 cells revealed HA-nsp3-4-V5 induced membrane alterations, consistent with an earlier report 293 for MERS-CoV and SARS-CoV 12 (Fig. 3, F and G). This included zippered ER and DMVs 294 with an average diameter of 145 nm. In contrast to HCV, the number of nsp3-4 induced 295 DMVs did not decrease in AGPAT single and double KO cells (Fig. 3G, left two panels). 296 However, in both cell pools we observed marked accumulations of multi-membrane vesicles 297 (MMVs), indicating the formation of aberrant membrane structures (Fig. 3, F and G).

298 To test whether similar to AGPAT1/2 relocalization to nsp3-4 induced DMVs, PA is also 299 enriched at those sites we used the GFP-tagged PA sensor derived from Raf1. In Huh7- 300 derived cells expressing SARS-CoV-2 HA-nsp3-4-V5, the functional version of the sensor 301 (GFP-PABD-Raf1-WT) strongly colocalized with HA-nsp3 in distinct puncta, whereas no 302 such puncta were found with the mutant PABD-Raf1, confirming specificity of PA sensor 303 recruitment to HA-nsp3-containing sites (Fig. 4, A and B). 304 Although in comparison to HCV, AGPAT1/2 DKO had lower impact on SARS-CoV-2 305 replication (compare Fig. 1F with Fig. 3E), and caused a morphologically distinct phenotype 306 of nsp3-4 induced DMVs (Fig. 2A and 3F, respectively), AGPATs, and most likely PA, still 307 accumulated at sites of SARS-CoV-2 DMV clusters (Fig. 4, A and B). This indicates that PA 308 synthesis pathways other than via AGPAT1/2, might contribute to SARS-CoV-2 replication 309 and DMV formation. By means of pharmacological inhibitors of enzymes that convert LPA, 310 PC and DAG to PA (fig. S7A), we measured the dose-dependent effect of these drugs on 311 SARS-CoV-2 replication. All inhibitors reduced SARS-CoV-2 replication in Calu-3 cells and 312 in A549 cells stably expressing ACE2, two commonly used cell models for this virus, at non- 313 cytotoxic concentrations, although in the case of the general AGPAT inhibitor CI976 314 selectivity was rather low (Fig. 4C and fig. S10A, respectively). Of note, combining the 315 inhibitors at concentrations close to or below their IC50 values caused much stronger 316 reduction of virus replication with no or minimal effect on cell viability, indicating that 317 SARS-CoV-2 can utilize PA produced by alternative PA synthesis pathways (fig. S10, A and 318 B). We then measured the effect of these drugs on PA accumulation at HA-nsp3 containing 319 puncta in HA-nsp3-4-V5 expressing cells and found that all inhibitors reduced PA levels at 320 these sites (Fig. 4D). This reduction was not the result of altered HA-nsp3-4-V5 expression 321 level or self-cleavage, which were unaffected in inhibitor-treated cells (fig. S10C). Next, we 322 determined if reduced PA levels caused by these inhibitors also affect SARS-CoV-2 nsp3-4 323 induced DMV formation. In cells treated with AGPAT, PLD1, and DAGK inhibitors DMV 324 diameters were significantly reduced (Fig. 4, E and F). Moreover, PLD2 inhibition promoted 325 the formation of MMVs and larger DMVs, similar to what we found in AGPAT single and 326 double KO cells (Fig. 3F). Taken together, our data suggest that PA enrichment is important 327 for proper SARS-CoV-2 DMV formation and viral replication. 328 Here, we show that PA produced by AGPAT1 and 2 is important for the replication of 329 evolutionary distant positive-strand RNA viruses, HCV and SARS-CoV-2 that amplify their 330 genome in association with DMVs. The remarkable dependence on a common host lipid for 331 the DMV biogenesis in these two viruses that differ profoundly in the diseases they cause and 332 in their biological properties, indicates a striking similarity in the biogenesis of these 333 organelles. Conversely, for viruses replicating their RNA genome in ER-derived membrane 334 invaginations such as the flaviviruses DENV and ZIKV, this lipid pathway appears to be 335 dispensable 4, 33. Of note, PA production through AGPAT1 and 2 is also involved in the 336 formation of autophagosome-like DMVs, arguing for some similarity between cellular and 337 viral DMV formation and lipid composition. Additionally, alternative routes of PA 338 biosynthesis contribute to HCV and SARS-CoV-2 replication and DMV generation. 339 At least three possibilities can be envisioned how PA promotes DMV formation in 340 viral replication and in the context of autophagy. First, the presence of lipids with cone or 341 inverted cone shape in membranes contributes to membrane bending by generating negative 342 or positive membrane curvature, respectively 16. While LPA has a large polar head group to 343 fatty acid tail ratio, giving rise to an inverse-cone shape and resulting in positive membrane 344 curvature, the additional fatty acid tail present in PA inverses the head-to-tail ratio. Hence PA 345 displays an overall cone shape, which contributes to negative membrane curvature. Thus, the 346 LPA - PA conversion by AGPATs might contribute to DMV formation by facilitating 347 membrane curvature. Second, PA is directly or indirectly implicated in membrane fission 34. 348 This might be achieved by recruitment of effector proteins by PA, either through downstream 349 signaling events, or directly by serving as docking site for PA-binding proteins that have

350 amphipathic or hydrophobic surfaces. In this regard, our NS4B-proteome showed the 351 enrichment of three known PA-interacting proteins, namely, Vitronectin, splicing factor-1, 352 and ubiquitin carboxy-terminal hydrolase L1, in the viral DMV fraction (data S1) 35. More 353 than 50 different proteins have been reported to interact with PA, including protein kinases, 354 phosphatases, nucleotide-binding proteins and regulators, however, a comprehensive list 355 remains elusive, and their possible role in the formation of DMVs during autophagy or viral 356 RNA replication, if any, remains to be determined 18. Third, an additional role of PA for the 357 functionality of viral replication organelles and perhaps also autophagosomes might be in 358 serving as an exchange lipid in a counter-transporter chain. In the case of HCV, we and 359 others identified accumulation of PI4P at DMVs 7 and similar findings have been made for 360 membranous structures involved in the early steps of autophagy 36. For HCV, it is thought 361 that PI4P recruits lipid transporters such as OSBP that deliver cholesterol into DMV 362 membranes in exchange for PI4P. A similar mechanism might apply for other lipids or the 363 PI4P precursor PI, with PA serving as a possible exchange factor against these other lipids or 364 PI, respectively. 365 The similar dependency of DMV-type replication organelles on PA, as reported here 366 for HCV and SARS-CoV-2, might offer an attractive starting point for broad-spectrum 367 antivirals targeting a diverse range of positive-strand RNA viruses replicating in such 368 structures. In line with this assumption, an inhibitor of cytosolic phospholipase A2α has been 369 reported to suppress replication and DMV formation of the 229E human coronavirus and to 370 exert antiviral activity also against the alphavirus Semliki forest virus 37. In addition, several 371 human diseases have been linked to defects in PA metabolism and selective autophagy, 372 including neurological disorders and chronic obstructive pulmonary disease 18, 19. Although 373 the precise role of PA in these diseases remains to be determined, the critical role of PA for 374 HCV and SARS-CoV-2 infection reported here might offer new approaches for therapeutic 375 intervention. 376 377 References and Notes: 378 1. W H O Situation Reports. Coronavirus disease (COVID-19) Weekly Epidemiological 379 Update and Weekly Operational Update.) (2021). 380 381 2. Spearman CW, Dusheiko GM, Hellard M, Sonderup M. Hepatitis C. Lancet 394, 382 1451-1466 (2019). 383 384 3. Wolf YI, et al. Origins and Evolution of the Global RNA Virome. mBio 9, (2018). 385 386 4. Neufeldt CJ, Cortese M, Acosta EG, Bartenschlager R. Rewiring cellular networks by 387 members of the Flaviviridae family. Nat Rev Microbiol 16, 125-142 (2018). 388 389 5. V'Kovski P, Kratzel A, Steiner S, Stalder H, Thiel V. Coronavirus biology and 390 replication: implications for SARS-CoV-2. Nat Rev Microbiol 19, 155-170 (2021). 391 392 6. Lamb CA, Yoshimori T, Tooze SA. The autophagosome: origins unknown, 393 biogenesis complex. Nat Rev Mol Cell Biol 14, 759-774 (2013). 394 395 7. Paul D, Bartenschlager R. Flaviviridae Replication Organelles: Oh, What a Tangled 396 Web We Weave. Annu Rev Virol 2, 289-310 (2015).

397 398 8. Romero-Brey I, et al. Three-dimensional architecture and biogenesis of membrane 399 structures associated with hepatitis C virus replication. PLoS Pathog 8, e1003056 400 (2012). 401 402 9. Cortese M, et al. Integrative Imaging Reveals SARS-CoV-2-Induced Reshaping of 403 Subcellular Morphologies. Cell Host Microbe 28, 853-866 e855 (2020). 404 405 10. Snijder EJ, et al. A unifying structural and functional model of the coronavirus 406 replication organelle: Tracking down RNA synthesis. PLoS biology 18, e3000715 407 (2020). 408 409 11. Wolff G, Melia CE, Snijder EJ, Barcena M. Double-Membrane Vesicles as Platforms 410 for Viral Replication. Trends Microbiol 28, 1022-1033 (2020). 411 412 12. Oudshoorn D, et al. Expression and Cleavage of Middle East Respiratory Syndrome 413 Coronavirus nsp3-4 Polyprotein Induce the Formation of Double-Membrane Vesicles 414 That Mimic Those Associated with Coronaviral RNA Replication. mBio 8, (2017). 415 416 13. Paul D, Hoppe S, Saher G, Krijnse-Locker J, Bartenschlager R. Morphological and 417 biochemical characterization of the membranous hepatitis C virus replication 418 compartment. J Virol 87, 10612-10627 (2013). 419 420 14. Yamashita A, et al. Glycerophosphate/Acylglycerophosphate acyltransferases. 421 Biology (Basel) 3, 801-830 (2014). 422 423 15. Takeuchi K, Reue K. Biochemistry, physiology, and genetics of GPAT, AGPAT, and 424 lipin enzymes in triglyceride synthesis. Am J Physiol Endocrinol Metab 296, E1195- 425 1209 (2009). 426 427 16. Kooijman EE, Chupin V, de Kruijff B, Burger KN. Modulation of membrane 428 curvature by phosphatidic acid and lysophosphatidic acid. Traffic 4, 162-174 (2003). 429 430 17. Zegarlinska J, Piascik M, Sikorski AF, Czogalla A. Phosphatidic acid - a simple 431 phospholipid with multiple faces. Acta Biochim Pol 65, 163-171 (2018). 432 433 18. Tanguy E, Wang Q, Moine H, Vitale N. Phosphatidic Acid: From Pleiotropic 434 Functions to Neuronal Pathology. Front Cell Neurosci 13, 2 (2019). 435 436 19. Mizumura K, Choi AM, Ryter SW. Emerging role of selective autophagy in human 437 diseases. Front Pharmacol 5, 244 (2014). 438

439 20. Fernandez-Galilea M, Tapia P, Cautivo K, Morselli E, Cortes VA. AGPAT2 440 deficiency impairs adipogenic differentiation in primary cultured preadipocytes in a 441 non-autophagy or apoptosis dependent mechanism. Biochem Biophys Res Commun 442 467, 39-45 (2015). 443 444 21. Cautivo KM, et al. AGPAT2 is essential for postnatal development and maintenance 445 of white and brown adipose tissue. Mol Metab 5, 491-505 (2016). 446 447 22. Lee JY, et al. Spatiotemporal Coupling of the Hepatitis C Virus Replication Cycle by 448 Creating a Lipid Droplet- Proximal Membranous Replication Compartment. Cell Rep 449 27, 3602-3617 e3605 (2019). 450 451 23. Khan I, et al. Modulation of hepatitis C virus genome replication by 452 glycosphingolipids and four-phosphate adaptor protein 2. J Virol 88, 12276-12295 453 (2014). 454 455 24. Zhang F, et al. Temporal production of the signaling lipid phosphatidic acid by 456 phospholipase D2 determines the output of extracellular signal-regulated kinase 457 signaling in cancer cells. Mol Cell Biol 34, 84-95 (2014). 458 459 25. Rizzo MA, Shome K, Watkins SC, Romero G. The recruitment of Raf-1 to 460 membranes is mediated by direct interaction with phosphatidic acid and is 461 independent of association with Ras. J Biol Chem 275, 23911-23918 (2000). 462 463 26. Prakash P, Hancock JF, Gorfe AA. Three distinct regions of cRaf kinase domain 464 interact with membrane. Sci Rep 9, 2057 (2019). 465 466 27. Sakane F, Hoshino F, Murakami C. New Era of Diacylglycerol Kinase, Phosphatidic 467 Acid and Phosphatidic Acid-Binding Protein. Int J Mol Sci 21, (2020). 468 469 28. Yamano K, et al. Endosomal Rab cycles regulate Parkin-mediated mitophagy. Elife 7, 470 (2018). 471 472 29. Vargas JNS, et al. Spatiotemporal Control of ULK1 Activation by NDP52 and TBK1 473 during Selective Autophagy. Mol Cell 74, 347-362 e346 (2019). 474 475 30. Holland P, et al. HS1BP3 negatively regulates autophagy by modulation of 476 phosphatidic acid levels. Nat Commun 7, 13889 (2016). 477 478 31. Knoops K, et al. SARS-coronavirus replication is supported by a reticulovesicular 479 network of modified endoplasmic reticulum. PLoS Biol 6, e226 (2008). 480

481 32. Wolff G, et al. A molecular pore spans the double membrane of the coronavirus 482 replication organelle. Science, (2020). 483 484 33. Welsch S, et al. Composition and three-dimensional architecture of the dengue virus 485 replication and assembly sites. Cell Host Microbe 5, 365-375 (2009). 486 487 34. Pagliuso A, et al. Golgi membrane fission requires the CtBP1-S/BARS-induced 488 activation of lysophosphatidic acid acyltransferase delta. Nat Commun 7, 12148 489 (2016). 490 491 35. Jang J-H, Lee CS, Hwang D, Ryu SH. Understanding of the roles of phospholipase D 492 and phosphatidic acid through their binding partners. In: Prog. Lipid Res.) (2012). 493 494 36. Judith D, Jefferies HBJ, Boeing S, Frith D, Snijders AP, Tooze SA. ATG9A shapes 495 the forming autophagosome through Arfaptin 2 and phosphatidylinositol 4-kinase 496 IIIbeta. J Cell Biol 218, 1634-1652 (2019). 497 498 37. Muller C, Hardt M, Schwudke D, Neuman BW, Pleschka S, Ziebuhr J. Inhibition of 499 Cytosolic Phospholipase A2alpha Impairs an Early Step of Coronavirus Replication in 500 Cell Culture. J Virol 92, (2018). 501 502

503 Acknowledgements 504 We thank Marie Bartenschlager, Lena Werstein, Ulrike Herian, Stephanie Kallis, Iris 505 Leibrecht and Fredy Huschmand for excellent technical assistance. We are grateful to Eliana 506 Acosta and Heeyoung Kim for editorial assistance. We acknowledge Alessia Ruggieri for 507 providing empty plasmid constructs. We also acknowledge the excellent support provided by 508 the Infectious Diseases Imaging Platform (IDIP) headed by Vibor Laketa, the University of 509 Heidelberg Electron Microscopy Core Facility (EMCF Heidelberg) headed by Stefan Hillmer 510 and the Proteomics and Metabolomics Facility (Pro-Met-) at CeMM. We thank the European 511 Virus Archive (EVAg) for the provision of the HP/F/2013 ZIKV strain. We also thank all 512 members of the Molecular Virology unit for continuous stimulating discussions. 513 Funding 514 This work was supported by grants from the Deutsche Forschungsgemeinschaft (DFG, 515 German Research Foundation) – Project Number 272983813 – TRR 179, Project Number 516 112927078 - TRR 83 and Project Number 314905040 – TRR 209, all to R.B. and Project 517 Number 112927078 - TRR 83, Project Number 240245660 - SFB1129, and Project Number 518 316659730 to B.B. V.P. is supported by a European Molecular Biology Organization 519 (EMBO) Long-Term Fellowship (ALTF 454-2020). C.J.N was supported in part by a 520 European Molecular Biology Organization (EMBO) Long-Term Fellowship (ALTF 466- 521 2016). L.K. and C.Z. acknowledge funding from the BMBF, grant number 031A602A 522 (ERASysApp SysVirDrug). P.V. and V.T. were supported by the Swiss National Science 523 foundation (SNF Project Number 310030_173085). G. S.-F. and K. H. were supported by a 524 European Research Council Advanced Investigator Grant (ERC AdG 695214 i-FIVE).

525

526 Contributions 527 Conceptualization, K.T., V.P., D.P., and R.B.; Investigation, K.T., V.P., D.P., J-Y.L., M-T.P., 528 W-I.T, C.J.N., M.C., B.C., C-S.T., C.L., P.V., K.H., A.C.M, C.Z., U.H., J.B., L.K., B.B.; 529 Writing – Original Draft, K.T., V.P., and R.B.; Writing – Review & Editing, K.T.,V.P., D.P., 530 J.Y.L., C.J.N., A.M, A.C.M., L.K., H.E., V.T., G.S-F., B.B. and R.B.; Funding Acquisition, 531 R.B., B.B., L.K. and G.S.-F. 532

533 Competing interests 534 Authors declare no competing interests. 535 536 Data and materials availability 537 All data is available in the main text or the supplementary materials. 538

539 Supplementary Materials: 540 Materials and Methods 541 Figures S1-S10 542 Tables S1-S5 543 References (1-30) 544 Other Supplementary Materials (Data S1-S3) 545 546

547 Figures 548

549 550 Fig. 1. Proteome analysis of HCV-induced DMVs identifies AGPATs as host 551 dependency factors critically contributing to viral replication. 552 (A) Experimental approach used to purify DMVs from HCV-replicating cells. (B) Volcano 553 plot of differentially enriched interactors of NS4B and calnexin (CNX). Q-values were 554 calculated using the limma software package and corrected for multiple hypothesis testing. 555 Viral proteins are highlighted with red letters. A magnified view with protein hits labeled is 556 given in fig. S1B. (C) A total of 139 genes were selected from the DMV proteome and 557 validated by siRNA screening (3 siRNAs per gene). CD81, PI4KA and Rluc were used as 558 positive controls; NC (negative control)1, NC2, GFP and mock infection served as negative 559 controls. A summary of the screening is given in Data S2. (D) Endogenous AGPAT1 and 2, 560 but not AGPAT3 are contained in NS4B-associated membranes. Membranes were purified

561 from naïve Huh7-Lunet cells (-), or Huh7-Lunet cells containing a subgenomic replicon 562 without or with an HA-tag in NS4B (NS4B-wt and NS4B-HA, respectively). Captured 563 proteins were analyzed by western blot, along with the input (2%). α-tubulin served as 564 loading control. (E) Colocalization of NS4B with AGPAT1 and 2. Huh7-Lunet cells stably 565 expressing AGPAT1- or AGPAT2-GFP were transfected with in vitro transcripts of the HCV 566 genome Jc1 and fixed 48 h post-transfection. (F) Effect of AGPAT KO on HCV replication. 567 Huh7.5 cells were infected with lentiviruses encoding AGPAT-targeting-sgRNA and 5 days 568 later, infected with an HCV reporter virus (JcR2a). After 48 h and 72 h, renilla luciferase 569 activities in cell lysates, reflecting viral RNA replication, were quantified. Graph shows 570 average and SD from 3 independent experiments. Significance was calculated by a paired t- 571 test. ***, p<0.001. Abundance of AGPAT proteins is shown on the bottom. α-tubulin served 572 as loading control. (G) Enzymatic activity of AGPAT is required for HCV replication. KO 573 cells were reconstituted with sgRNA-resistant AGPAT wild-type (wt) or catalytically dead 574 mutants (M1 and M2) by lentiviral transduction. Cells were infected with JcR2a, and renilla 575 luciferase activities were quantified. Graph shows average and SD from 3 independent 576 experiments. Significance was calculated by paired t-test. ***, p<0.001. ns, p>0.05. Note the 577 complete rescue by AGPAT1 and 2 co-expression. Abundance of AGPAT proteins is shown 578 below the graph; α-tubulin served as loading control. (H) AGPAT1/2 DKO does not affect 579 DENV or ZIKV propagation. Cells were infected with DENV-2 (strain 16681) or ZIKV 580 (strain H/PF/2013) and 48 h later virus titer was quantified by plaque assay. Graph shows the 581 average and SD from 3 independent experiments. Significance was analyzed by a paired t-test. 582 ns, p>0.05. PFU, plaque forming units.

583 584 585 Fig. 2. Requirement of AGPATAs for HCV-induced DMV formation and PA 586 accumulation on HCV-induced DMVs and autophagy-related structures. 587 (A to C) AGPAT1/2 DKO dampens DMV formation induced by HCV. Huh7-derived cells 588 stably expressing the T7 RNA polymerase and containing or not a double knock-out (DKO)

589 of AGPAT1 and 2 were transfected with a HCV replicase-encoding plasmid containing a 590 GFP insertion in NS5A (construct pTM NS3-3’/5A-GFP, top panel). Transcripts are 591 generated from the plasmid in the cytoplasm via the T7 promoter and terminator (T7-term) 592 sequence and the HCV NS3 – 5B coding region is translated via the IRES of the 593 encephalomyocarditis virus (EMCV). (A) After 24 h, cells were fixed and subjected to 594 CLEM. Low resolution confocal microscopy images identifying transfected cells are shown 595 on the top left. The area in the red box is shown in the corresponding EM image. Yellow 596 arrow heads indicate DMVs. Insets at the bottom indicate zoomed-in regions. (B) DMVs 597 within whole cell sections were counted and divided by cell area (µm2). Graph shows average 598 and SD from 4 different transfected cells. Cells expressing comparable level of HCV 599 replicase were selected for EM analysis. Significance was calculated by a paired t-test. **, 600 p<0.01. (C) Expression levels of NS4B and NS5A in transfected cells were determined by 601 western blotting. (D) Lipidome analysis of HCV-induced DMVs. Extracts of Huh7 cells 602 containing the subgenomic replicon sg4BHA31R (NS4B-HA) and Huh7 cells stably 603 overexpressing HA-tagged Calnexin (CNX-HA) and control Huh7 cells were prepared as 604 described in supplementary methods and used for HA-affinity purification under native 605 conditions. An aliquot of the sample was analyzed by electron microscopy (top panels) 606 whereas the majority was subjected to lipidome analysis by using mass spectrometry. Values 607 obtained for the NS4B-HA sample were normalized to those obtained for the CNX-HA 608 sample that was set to one. The complete list of analyzed lipids is summarized in data S3. (E) 609 PA accumulation at NS5A containing structures. Huh7-Lunet/T7 cells were transfected with 610 a construct analogous to the one in panel A, but containing a mCherry insertion in lieu of 611 GFP, along with an EGFP-tagged wildtype (WT) or mutant (4E) PA sensor (construct pTM- 612 EGFP-PABD-Raf1-WT or -4E). Twenty-four hours later, GFP-PABD and NS5A-mCherry 613 were visualized by fluorescence microscopy. White boxes indicate regions magnified in the 614 lower right of each panel. (F) Top panel: Alternate PA biosynthesis pathways via 615 lysophosphatidic acid (LPA) or phosphatidylcholine (PC) catalyzed by AGPATs or PLDs, 616 respectively. Bottom panel: Huh7-Lunet/T7 cells were electroporated with in vitro transcripts 617 of a subgenomic HCV reporter replicon encoding the firefly luciferase. Four hours after 618 transfection, different concentrations of PA synthesis inhibitors were added to the cells and 619 luciferase activities were analyzed at 48 h after electroporation. Graph shows average and SD 620 from 3 independent experiments. Cell viability determined by CellTiter-Glo luminescent 621 assay is indicated with the red line. (G) PA recruitment to autophagy-related structures in 622 selective and non-selective autophagy. Top panel: Huh7-derived cells expressing EGFP- 623 PABD-Raf-1 were incubated in growth medium (top left panels) or in serum-free medium 624 with 200 nM BafA1 (top right panels) for 3 h. Cells were fixed and stained with a LC3 625 specific antibody. Bottom panel: For selective autophagy, mCherry-tagged Parkin was co- 626 expressed with EGFP-PABD-Raf1, followed by incubation with 10 µM Valinomycin to 627 induce mitophagy. Cells were fixed after 3 h, and GFP-PABD and mCherry-Parkin were 628 visualized by fluorescence microscopy. Images in panels E and G are maximum intensity 629 projections. 630 631

632 633 Fig. 3. AGPATs are recruited to SARS-CoV-2 induced DMVs and contribute to viral 634 replication 635 (A) Change of subcellular localization of AGPATs upon expression of SARS-CoV-2 nsp3-4. 636 Huh7-derived cells transiently expressing AGPAT2-GFP were transfected with a SARS- 637 CoV-2 HA-nsp3-4-V5 expression construct or the empty vector. After 48h, cells were stained

638 with HA-specific antibody and examined by confocal microscopy. Maximum intensity 639 projections are shown. Enrichment score indicates the likelihood of cells showing a punctate 640 or diffuse staining pattern. (B) Clustering of AGPAT2-GFP in SARS-CoV-2 HA-nsp3-4-V5 641 expressing cells. Huh7-Lunet/T7 cells were co-transfected with AGPAT2-GFP and SARS- 642 CoV-2 HA-nsp3-4-V5 or the empty vector. Twenty-four hours later, cells were fixed and 643 ~1000 cells per condition were separated into two morphotypes (diffuse or punctate) using 644 CellProfiler Analyst based semi-supervised classifier. Significance was calculated using an 645 unpaired t-test. *, p<0.05. (C) AGPAT clustering occurs independent of ER remodeling 646 induced by nsp3-4. Huh7-Lunet cells expressing AGPAT2-GFP and HA-nsp3-4-V5 were 647 stained for the ER markers protein disulfide isomerase (PDI) and calnexin and analyzed by 648 confocal microscopy. (D) AGPATs are localized at SARS-CoV-2 HA-nsp3-4-V5 induced 649 DMVs. Huh7-derived cells were transiently transfected with AGPAT2-GFP HA-nsp3-4-V5 650 and subjected to CLEM. Light and EM images were correlated by using lipid droplets as 651 fiducial markers. White and yellow boxes indicate areas magnified in the corresponding 652 panels on the right. (E) AGPAT1/2 contribute to SARS-CoV-2 replication. Huh7-Lunet 653 control, AGPAT2 single (SKO) and AGPAT1/2 double (D)KO cells were infected with 654 SARS-CoV-2 (MOI=0.1). Twenty-four hours later, cells were fixed and immunostained for 655 nucleocapsid, and the percentage of N-positive cells was determined using CellProfiler. 656 Normalized data from three biologically independent experiments are plotted (top right 657 panel). Total RNA was isolated from infected cells, and SARS-CoV-2 RNA levels were 658 determined using RT-qPCR. Data were normalized to cellular GAPDH mRNA (bottom right 659 panel). Significance was calculated using ordinary one-way ANOVA. *, p<0.05. (F) Aberrant 660 SARS-CoV-2 DMVs in AGPAT1/2 DKO cells. Huh7-Lunet cells with single (SKO) or 661 double knock-out (DKO) and stably expressing T7 polymerase were transfected with a 662 plasmid encoding SARS-CoV-2 HA-nsp3-4-V5 and fluorescent neon-green. Twenty-four 663 hours later, cells were fixed and NeonGreen positive cells were recorded and examined by 664 EM. HA-nsp3-4-V5 induced DMVs and multi-membrane vesicles (MMVs) were quantified. 665 Shown are the number and diameter of DMVs and MMVs in these cells as observed from at 666 least 8 cell profiles per condition. Statistical significance was calculated using ordinary one- 667 way ANOVA. ****, p<0.001. Light microscopy images in panels A to D are maximum 668 intensity projections. 669 670 671 672

673 674 675 Fig. 4. PA accumulation at SARS-CoV-2 DMVs and role of alternative PA synthesis 676 pathways for SARS-CoV-2 replication and DMV formation. (A) PA enrichment at 677 SARS-CoV-2 nsp3-containing structures. Huh7-Lunet cells expressing the wildtype or 678 mutant form of the PA sensor were transfected with the plasmid encoding HA-nsp3-4-V5 and

679 24 h later, cells were fixed, immunostained with HA-specific antibody and HA-nsp3 and 680 GFP-PABD were visualized by confocal microscopy. Maximum intensity projections are 681 shown. (B) Using CellProfiler Analyst, a semi-supervised machine learning classifier was 682 trained to differentiate between punctate and diffuse signals of the GFP-PABD sensor (top 683 panel). A normalized enrichment score which indicates the probability of cells showing 684 punctate GFP-PABD localization to nsp3 fluorescent signal across the whole cell population 685 is shown in the graph on the bottom panel. Significance was calculated by unpaired t-test. 686 ****, p<0.0001. (C) Alternative pathways for PA generation are important for SARS-CoV-2 687 replication. Calu-3 cells were infected with SARS-CoV-2 (MOI=5) in the presence of 688 AGPAT, PLD1/2, or DAGK inhibitors. Cells were fixed 24 h post infection, stained with 689 nucleocapsid-specific antibody and percentage of infected cells was quantified using 690 CellProfiler. Cell viability and percentage inhibition are plotted as dose-response curves and 691 IC50 values are given on the top of each panel. (D) AGPAT, PLD and DAGK inhibitors 692 reduce PA accumulation at nsp3-positive structures. Huh7-Lunet cells were transfected with 693 SARS-CoV-2 HA-nsp3-4-V5 and GFP-PABD-Raf1 encoding plasmids, followed by addition 694 of a given inhibitor 4h after transfection. Twenty-four hours later, cells were fixed and HA- 695 nsp3 was detected with an HA-specific antibody. GFP-PABD and HA-nsp3 were visualized 696 by confocal microscopy. A semi-automated machine learning based classifier was trained to 697 separate HA-nsp3/PABD double-positive structures from HA-nsp3 single positive structures. 698 Enrichment score for HA-nsp3/PABD double-positive structures showing the up or 699 downregulation of double positive cells in different samples is plotted and statistical 700 significance was calculated using ordinary one-way ANOVA. *, p<0.05, **, p<0.005. (E) 701 Decrease of SARS-CoV-2 DMV diameter by AGPAT, PLD and DAGK inhibitors. Huh7- 702 Lunet/T7 cells were transfected with the plasmid encoding HA-nsp3-4-V5 and fluorescent 703 NeonGreen, followed by addition of inhibitors 4 h after transfection. Twenty-four hours later, 704 cells were fixed, NeonGreen positive cells were recorded and examined by EM. 705 Representative images are shown for each condition. (F) Number and morphology of DMVs 706 were determined for at least 7 cell profiles per condition. DMV diameters are plotted and 707 statistical significance was calculated using ordinary one-way ANOVA. ****, p<0.001.